The purpose of this project is to train a neural network that predicts the steering angle of a car simulator as it traverses a course. This is done not by using deterministic rules but rather by cloning behaviour. Data is collected by driving the car simulator around the course. A network is then trained on this data. A successful outcome is achieved when the car traverses the course in autonomous mode without it ever leaving the course.
The car simulator is provided by Udacity along with a python program that can send the steering angles (and speed) to the simulator. The simulator provides a image captured by a dash camera that the trained network will then use to predict the correct steering angle.
Here is a gif showing you hwo the manual simulator works: